Time series prediction model based on transformer and LSTM for predicting the occurrence rate of mountain torrents

With the acceleration of global climate change and urbanization, the frequency and scope of mountain torrents continue to expand, posing a serious threat to human life and property safety. Predicting the occurrence time and impact range of mountain torrents disasters in advance, evaluating the risk of mountain torrents disasters, can help prevention and control departments make decisions, conduct early warning and governance, and is of great significance for disaster prevention and reduction. Therefore, the paper proposes a time-series prediction model based on Transformer and LSTM for predicting the disaster rate of mountain torrents. Firstly, taking the Longnan region of Gansu Province in China as an example, the paper selected 31 influencing factors and used PCA algorithm for dimensionality reduction. An 8-dimensional temporal vector was selected as the input of the Transformer model, and the attention mechanism of the Transformer was used for encoding and decoding. Then, LSTM was input for temporal prediction of mountain torrents disasters. A series of experiments have shown that the algorithm studied in this paper can achieve a testing accuracy of 0.9372 and a validation accuracy of 0.8629, which is higher than the prediction efficiency of traditional machine learning models. The AUC index is also the best among all models. The paper innovatively presents a prediction model for the occurrence rate of mountain torrents disasters, which has good application prospects and promotion value. After application, it can provide decision support for relevant departments.